the observations used to study climate fluctuations are derived from observing programs that were developed not for this purpose, but rather to support day-to-day weather forecasting. Since the day-to-day variability of the atmospheric system in much of the world is far larger than the decadal climate variability, identification of multi-decadal climate fluctuations is often hampered by the poor quality and lack of homogeneity of the observations. For example, Groisman and Easterling (1995), in their study of the variability and trends of precipitation over North America, painstakingly document the many discontinuities and false jumps in the climate record that arise from changes in observing practices. These changes, if ignored, can and do overwhelm important long-term precipitation fluctuations.
Other atmospheric quantities besides precipitation are also affected by problems in measurement. In this section, Robinson (1995) points out the large disparity in snow-cover extent between a NASA data set using microwave measurements and a NOAA data set using visible imagery. Karl et al. (1995) devote considerable discussion to jumps and trends introduced into the surface temperature record by such factors as urban heat islands, changes in irrigation practices, differences in instrumentation, and station relocations. Cayan et al. (1995) use surface marine observations to calculate the latent and sensible heat fluxes that are used as forcing agents of the sea surface temperature field in the North Pacific. Their analysis includes the basin-wide climate jump that began about 1976-1977. To account for known systematic biases related to trends in the wind field and the sea surface and marine air-temperature and moisture lapse rates (Ward, 1992; Cardone et al., 1990; Wright, 1988; Ramage, 1987), the global trend of each of these quantities is removed. As mentioned in Zebiak's commentary on Cayan's paper, it is unfortunate that such adjustments to the data are necessary.
Often the corrections and adjustments that must be applied to a climate record are of such magnitude that it is difficult to be confident that the resulting time series adequately reflects the climate fluctuations. For example, Figure 2 illustrates the significant adjustments required to calculate global temperature fluctuations since the nineteenth century. Through the use of other data bases (e.g., changes in snow cover, alpine glaciers, or sea level) and the isolation of various components of the surface temperature record (e.g., marine air temperature, sea surface temperature, and land temperature), it is possible to gain more confidence in the adjustments applied. Each of the data sets has distinctly different problems related to long-term homogeneity and data quality that require independent adjustment procedures. When these independent data sets provide a physically consistent picture of decade-to-century climate fluctuations, their agreement can be very compelling. In fact, the analysis of quasi-periodic oscillations of surface winds, pressure, and ocean and marine air temperatures
by Deser and Blackmon (1995) relies entirely on physical consistency among independent variables.
In order to improve our ability to discern climate fluctuations on decade-to-century time scales within the existing climate record, some federal agencies such as NSF, DOE, and NOAA are supporting data archeology efforts. Data archeology is the process of seeking out, restoring, correcting, and interpreting data sets. Such efforts are critical to identifying and understanding longer-scale climate fluctuations, since they often turn up information that reveals a pattern not otherwise obvious.
The Comprehensive Ocean-Atmosphere Data Set (COADS), which is just one of several major data-archeology efforts, provides a good example of the type of benefits that can be expected from such efforts. Now, several years after the project's inception (Woodruff et al., 1987), scientists are identifying decadal-scale variations and changes in many climate elements previously thought to be too uncertain to document with any confidence (London et al., 1991; Flohn et al., 1990; Parungo et al., 1994).
Interestingly, up to the present, satellite data have not figured prominently in the analysis of multi-decadal climate variability. Their short observing history and a lack of temporal homogeneity have hampered efforts to use them. Certainly, there are notable exceptions, such as NOAA's snow and sea-ice products and the Spencer and Christy (1990) microwave sounding-unit data. However, major temporal inhomogeneities in data sets, like those identified in the International Satellite Cloud Cover Project (ISCCP; see Klein and Hartmann, 1993), threaten the usefulness of much of the multi-decadal satellite data. The two major challenges over the next few decades for the United States will to be to ensure that global-change satellite-research projects such as EOS provide homogeneous data over their planned life-